Games p2p social p2p some historical observations
1 / 48

Games & P2P->Social->P2P :- Some historical observations - PowerPoint PPT Presentation

  • Uploaded on

Games & P2P->Social->P2P :- Some historical observations. Jon Crowcroft, Feb 23 2012. Outline. Brief History of VR, P2P, OSNs Two Main Tech. Items Real Life and Virtual Space/Affinity P2P Capacity Lessons for Games etc AOB. What I do…. I don’t play games

I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
Download Presentation

PowerPoint Slideshow about ' Games & P2P->Social->P2P :- Some historical observations' - yates

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.

- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript
Games p2p social p2p some historical observations

Games & P2P->Social->P2P :-Some historical observations

Jon Crowcroft,

Feb 23 2012


Brief History of VR, P2P, OSNs

Two Main Tech. Items

Real Life and Virtual Space/Affinity

P2P Capacity Lessons for Games etc


What i do
What I do…

I don’t play games

I don’t download of P2P systems

I don’t like Online Social Networks

I’ve done extensive research on both

Since 1981…I’ve mostly mucked about with IP

Before the internet
Before the Internet

  • Lots of proprietary, closed nets

  • And Usenet (UUCP ++ )

  • Muds, and Moos and Usenet

  • HLA – first go at multicast/p2p games

  • Napster and Freenet

  • MMORPG – mostly Big Iron Server

  • Facebook, Peerson, Diaspora – P2P OSNs

  • Angry Birds and Ed Milliband

  • Trust, Clouds and Fiber to the Home

1 john l miller and jon crowcroft from at netgames 2009

1. John L. Miller and Jon Crowcroft

from at NetGames 2009



  • Lots of Online Game / DVE research proposing new message propagation models

  • Typically evaluated against synthetic workloads

    • How do these compare to real workloads?

  • Most DVE users play World of Warcraft (WoW)

    • Battlegrounds are a tractable, dynamic scenario


  • Based in a fantasy environment: knights and wizards…

  • Avatars organized into two teams: ‘factions’

  • Compete over resources or objectives

    • Dominate combat and geography

    • Mixtures of melee and spell/missile combat

  • Battle duration: ~5 to ~30 minutes

  • Battle participants: 10 - 240

  • Both sides rewarded, winner > loser


10-30 players

Control stationary flags

First team to 1600 wins


7 yards/s Running

14 yards/s Riding


~5 yard Melee range

30 (45) yard Spell range

~500 yard visual range

500 yards



Battle Excerpt Video

Abstracted Moves (8x speed)


  • Capture data using Windows Network Monitor 3.3

  • Custom move extraction library

    • Parse .cap files into TCP payloads

    • Process payloads and extract movement data

    • Output .csv movement trace

  • Gather landmark data

    • Join battle, circle around landmarks 


  • Join battleground with two grouped Avatars

    • Ensures they join the same battleground

  • Move to opposite ends of the map, stealth

  • Try not to fight or die

    • Team-mates don’t like this

  • Save resulting capture, filter observers


  • Analyzed 13 Battles

    • Scores from 1600-0 to 1600-1590

    • Observer team won 6, lost 7

  • 392 unique avatars, 456 avatar instances

  • Average avatar play interval: 69% of battle

  • Average data continuity: 73% of interval


  • Expecting:

    • Hotspots. Avatars spend most of their time concentrated in a few common areas

    • Waypoint navigation. Avatars move along well-defined paths to well-defined destinations

    • Grouping. Avatars move together to their destination

      • Avatars start together: this should be a no-brainer


  • Determine hotspots by counting seconds spent at each location in the battleground

    • Divide battleground into a grid

    • Sum avatar seconds spent in each cell

    • Cells with highest count are hotspots for that battle

  • Hotspots were found where expected, but not in every battle

    • Hotspots typically at flags and graveyards

  • Some hotspots on heavy travel paths: ambush!

  • Top five hotspots vary battle to battle


  • Waypoint movement should follow fixed paths

  • Movement geographically constrained

    • Avoid water, which slows to 25% of riding speed

    • Cliffs / hills / rivers channel movement

  • We found many paths used between hotspots

    • ‘Patrollers’ (16% of avatars) follow waypoints

    • ‘Guards’ (12%) move around a preferred area

    • ‘Wanderers’ (49%) move throughout the map

    • (23% of avatars observed too little to classify)

  • Waypoints useful, but not sufficient


  • Logically, Avatars should stick together

    • They start together, and resurrect together

    • Outnumber the enemy to stay alive

  • In fact, they seem to go out of their way NOT to stick together

  • Analysis: sum up all player seconds where avatar is within 30 yards of another avatar

    • Ideally, should include movement requirement, but this is a much looser / more generous metric.

Affinity Map

Non-Affinity Map



  • Existing Avatar movement models insufficient

    • Hotspots useful, but not consistent

    • Waypoints useful for a (small) subset of avatars

    • Grouping / flocking useful for a minority of avatars

  • A new synthetic movement model is needed

    • In the meantime, use real data

Now what about social nets
Now what about Social Nets

  • Lets look at movement in the real world

  • 4 examples –

    • Conference

    • Building

    • Disaster

    • Epidemic


Pittman / GauthierDickey: “Measurement Study of Virtual Populations” (WoW Census+)

Suznjevic et. al. “Action specific MMORPG traffic analysis: Case study of World of Warcraft”

Svoboda et. al. “Traffic Analysis and Modeling for World of Warcraft” (mobile packet traces)

Thawonmas et. al. “Detection of Landmarks for Clustering of Online-Game Players” (ICE / Angel’s Love)

Chen and Lei – “Network game design: hints and implications of Player Interaction” (ShenZou network traces)

La and Michiardi – “Characterizing user mobility in second life”

Liang et al. – “Avatar Mobility in Networked Virtual Environments: Measurements, Analysis, and Implications” (second life)


  • Further analysis of network traces

    • Message attribution

  • Simulate proposed DVE architectures

    • Client-server, application-layer multicast, mesh

    • Aggregation / per message transmission

  • Capture Wintergrasp data

    • Most challenges are practical, not technical

  • Contact me for access to anonymized traces: [email protected]

Source: Woodcock)


The near term in feasibility of p2p mmog s

The Near-Term (in)Feasibility of P2P MMOG’s

2. John L. Miller and Jon Crowcroft

From NetGames 2010


  • Motivation

  • Data Capture and Processing

  • Operational Assumptions

  • Simulator

  • Results

  • Conclusions

Paper motivation
Paper Motivation

  • Challenges in DVE’s well known

    • Scalability, latency, security

  • Lots of great proposals, especially P2P

    • No significant adoption. Why?

  • Search real examples for answers

    • World of Warcraft or Second Life

    • Realistic network conditions

      • Everything works differently in the real world

Data acquisition
Data Acquisition

  • Getting network traces is hard

    • Players are (rightly) paranoid

    • Even online affiliation provided little incentive

  • Interpreting network traces is hard

    • WoW protocol is pithy and partly secured

    • Mitigation: internet is a treasure trove of information

  • End result: simulation input traces which include:

    • Avatar position, movement, and some activities

    • ‘Attribution’ of most message bytes which were not successfully parsed

    • Non-parsed, non-attributed bytes discarded

Data parsing results
Data Parsing Results

Categories as proposed in earlier research, e.g. Suznjevic, Dobrijevic, Matijasevic, 2009

Operational assumptions
Operational Assumptions

  • Residential players with typical broadband

    • Actual, not advertised speeds

  • Messages

    • Originate at ‘sender’ Avatar position

    • Transmitted by originator to each receiver

  • Propagation determined by Area of Interest (AoI)

    • (Optimistic) perfect, zero-cost knowledge available

  • Transport is TCP/IP

    • Guaranteed once, in-order message delivery


  • Accepts our WoW simulation traces as input

  • Time resolution: 1 millisecond

  • Includes modelling of

    • TCP Windowing, packet framing

    • Last hop uplink, downlink capacity and contention

    • Latency

  • Node bandwidth assignments based upon OfCom 2009 UK survey


  • Even without overhead, P2P protocols consume too many resources

    • For intense scenarios, nodes can ‘fall behind’

    • Most messages important and difficult to discard

  • ‘Falling behind’ increases node load

    • new messages + catch-up

  • Aggregation of messages useful

    • Introduces sender delay

    • Offset by fewer bytes to transmit

Results bandwidth
Results: Bandwidth

  • P2P Upload bandwidth scales with number of peers in AoI, 10 x to 100 x the Client-server load on average

  • Client-server Download slightly larger because nearby AI is communicated from server rather than locally calculated

Results latency
Results: Latency

  • Latency is usually a function of message creation, upload bandwidth, and neighbors

  • Average latency OK with small number of neighbors, but as expected, scenarios with large number of neighbors and high message rates have highest latency

  • Differences from Client: server: Ignoring questing, 40:1 to 445:1

Message aggregation
Message Aggregation

  • Message aggregation: bundling together time-proximate messages to send together

    • Reduces overhead by up to a factor of 10

  • Delays message transmission, but mitigated by shorter transmission time and smaller transmission queue

  • Note: some ‘bundling’ artifacts in traces

    • Many message types lack internal timestamp, so we used capture time as timestamp

    • Wow client-server does some aggregation, biases our measurement with bundles of messages

P2p message aggregation
P2P Message Aggregation

Impact of 1 millisecond of aggregation on average latency measurements, relative to original measurements

Client server aggregation
Client-Server Aggregation

Impact of 1 millisecond of aggregation on average latency measurements, relative to original measurements


  • P2P wow-like MMOG’s are not feasible with today’s residential broadband

    • Hybrid solutions may be possible

  • Message aggregation useful

    • Reduce bandwidth and in many cases latency

  • Future work:

    • Recommend focus on non-P2P solutions

    • Identify ‘average’ node attributes required to support P2P

How do p2p osns do
How do P2P OSNs do?

  • 3 main attempts

  • Diaspora, Peerson and Safebook

  • Complexity is high

  • Reliability is low

  • Downlink capacity not, yet, a problem, compared to games

  • Gamers (FPS) have to cope with field of view/realestate, so natural limit on total players

  • C.f. regions in first part of talk, gives scale

  • Would same region idea work for OSNs

  • i.e. spatial affinity in real world, for relevance

  • Research (out there recently) says YES!

Take home
Take Home

We’re not there yet…

Server based systems for many-to-many Games and OSNs

+ve ordering/cheat proofing

+ve mix/filter

-ve scale of server

-ve control/privacy

With fiber roll-out happening, this will work (as well as lower latency)

Using Affinity for relevance filtering in OSN update traffic could work too